考虑残差学习的深层卷积神经网络在结构损伤识别中的应用研究

李书进; 熊书琪; 范沛然; 王罡

doi:10.13204/j.gyjzg21101009

考虑残差学习的深层卷积神经网络在结构损伤识别中的应用研究

doi: 10.13204/j.gyjzg21101009

1. 武汉理工大学土木工程与建筑学院, 武汉 430070;
2. 中冶建筑研究总院(深圳)有限公司, 广东深圳 518054

基金项目:

国家自然科学基金项目（52078399）。

详细信息

作者简介:
李书进,男,1967年出生,博士,教授。电子信箱:sjli@whut.edu.cn

计量
- 文章访问数: 147
- HTML全文浏览量: 25
- PDF下载量: 8
- 被引次数: 0
出版历程
- 收稿日期: 2021-10-10
- 网络出版日期: 2022-10-28

Application Research on Deep Convolutional Neural Network Considering Residual Learning in Structural Damage Identification

1. School of Civil Engineering and Architecture, Wuhan University of Technology, Wuhan 430070, China;
2. Central Research Institute of Building and Construction(Shenzhen) Co., Ltd., Shenzhen 518054, China

摘要

摘要: 提出了一种考虑残差学习的深层卷积神经网络损伤识别方法，并将其应用到框架结构节点损伤识别中。采用试验研究方式对所提方法进行了深入探讨，结果表明该方法可以很好地解决网络深化带来的网络退化或梯度爆炸、弥散导致的收敛困难和识别准确率差等问题，能对结构损伤诊断中的损伤定位这一复杂问题进行有效识别。在对试验框架节点损伤位置识别的对比研究中，考虑残差学习的深层卷积神经网络收敛速度和准确率均高于常规浅层神经网络和深层神经网络，有极高的准确率和稳定性，从而使得对于工程中复杂结构损伤诊断所需要的更深层、更复杂网络的搭建成为可能。此外，为提升网络用训练样本的质量和数量，依据样本划分规律提出了一种新的数据样本扩增方法，该方法在相同条件下可以显著增加用以训练的样本量并能弱化数据截断带来的信息缺失，识别准确率和收敛速度也大幅提高，研究显示了该处理方式的有效性和适用性。
- 损伤识别 /
- 卷积神经网络 /
- 试验研究 /
- 工程结构 /
- 残差学习
Abstract: A deep convolutional neural network damage identification method considering residual learning was proposed and applied to the damage identification of the frame structure joints. The proposed method was deeply discussed by means of experimental research, and the results showed that this method could solve the problems of convergence difficulty and poor recognition accuracy caused by the network degradation and gradient explosion, dispersion problems when the network deepening. In the comparative study of joint damage identification of test frame, the convergence speed and accuracy of deep convolutional neural network considering residual learning were higher than those of shallow conventional neural network and deep neural network, and had high accuracy and stability, and increased the possibilities to build a deeper and more complex network for damage diagnosis of complex structures in engineering. In addition, in order to improve the quality and quantity of training samples for network, a new data processing method was proposed according to the law of sample division. This method could significantly increase the sample size for training, weaken the information loss caused by data truncation under the same conditions, and greatly improve the recognition accuracy and convergence speed, and the research showed its effectiveness and applicability.
- damage identification /
- convolutional neural network /
- experimental study /
- engineering structure /
- residual learning

HTML全文

参考文献(21)

[1]	黄新波, 胡潇文, 朱永灿, 等. 基于卷积神经网络算法的高压断路器故障诊断[J]. 电力自动化设备, 2018, 38(5):136-140.
[2]	吴春志, 江鹏程, 冯辅周, 等. 基于一维卷积神经网络的齿轮箱故障诊断[J]. 振动与冲击, 2018, 37(22):51-56.
[3]	王丽华, 谢阳阳, 周子贤, 等. 基于卷积神经网络的异步电机故障诊断[J]. 振动、测试与诊断, 2017, 37(6):1208-1215.
[4]	李雪松, 马宏伟, 林逸洲. 基于卷积神经网络的结构损伤识别[J]. 振动与冲击, 2019, 38(1):159-167.
[5]	ZHANG Y, MIYAMORI Y, MIKAMI S, et al. Vibration-based structural state identification by a 1-Dimensional convolutional neural network[J]. Computer-Aided Civil and Infrastructure Engineering, 2019, 34(9):822-839.
[6]	李书进, 赵源, 孔凡, 等. 卷积神经网络在结构损伤诊断中的应用[J]. 建筑科学与工程学报, 2020, 37(6):29-37.
[7]	GU J, WANG Z, KUEN J, et al. Recent advances in convolutional neural networks[J]. Pattern Recognition, 2018, 77:354-377.
[8]	胡卫兵, 杨佳, 王龙, 等. 基于层合理论和误差反向传播神经网络的古建筑木结构损伤识别和量化研究[J]. 工业建筑, 2020, 50(11):71-77.
[9]	朱宏平, 张源. 基于自适应BP神经网络的结构损伤检测[J]. 力学学报, 2003, 35(1):110-116.
[10]	YU D, DENG L. Deep Learning and its applications to signal and information processing[J]. IEEE Signal Processing Magazine, 2011, 28(1):145-154.
[11]	KRIZHEVSKY A, SUTSKEVER I, HINTON G. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6):84-90.
[12]	REHMER A, KROLL A. On the vanishing and exploding gradient problem in gated recurrent units[J]. IFAC Papersonline, 2020, 53(2):1243-1248.
[13]	HADSELL R, RAO D, RUSU A A, et al. Embracing change:continual learning in deep neural networks[J]. Trends in Cognitive Sciences, 2020, 24(12):1028-1040.
[14]	LAAVANYA M, VIJAYARAGHAVAN V. Residual learning of transfer-learned AlexNet for image denoising[J]. IEIE Transactions on Smart Processing & Computing, 2020, 9(2):135-141.
[15]	HUBEL D H, WIESEL T N. Receptive fields, binocular interaction and functional architecture in the cat's visual cortex[J]. The Journal of Physiology, 1962, 160(1):106-154.
[16]	GHOSH A, KUMAR H, SASTRY P. Robust loss functions under label noise for deep neural networks[C]//AAAI. 31st AAAI Conference on Artificial Intelligence. Palo Alto:2017:1919-1925.
[17]	VEIT A, WILBER M J, BELONGIE S. Residual networks behave like ensembles of relatively shallow networks[J]. Advances in Neural Information Processing Systems, 2016, 29:550-558.
[18]	董聪, 丁辉, 高嵩. 结构损伤识别和定位的基本原理与方法[J]. 中国铁道科学, 1999, 20(3):89-94.
[19]	KIM J T, RYU Y S, CHO H M, et al. Damage identification in beam-type structures:frequency-based method vs mode-shape-based method[J]. Engineering Structures, 2003, 25:57-67.
[20]	HE K, ZHANG X, REN S, et al. Delving deep into rectifiers:surpassing human-level performance on ImageNet classification[C]//IEEE. International Conference on Computer Vision. New York:2015:1026-1034.
[21]	MANCEV D, TODOROVIC B. A primal sub-gradient method for structured classification with the averaged sum loss[J]. International Journal of Applied Mathematics & Computer Science, 2014, 24(4):917-930.